Plasma Proteomics Identifies a Chemokine Storm in Idiopathic

Received: 6 April 2018 | Revised: 12 April 2018 | Accepted: 17 April 2018 DOI: 10.1002/ajh.25123

RESEARCH ARTICLE AJH

Plasma proteomics identifies a ‘chemokine storm’ in idiopathic multicentric Castleman disease

1 University of Pennsylvania, Philadelphia, Human Herpesvirus-8 (HHV-8)-negative/idiopathic multicentric Castleman disease (iMCD) is a Pennsylvania; 2Hospital of the University of poorly understood disease involving polyclonal lymphoproliferation with dysmorphic germinal cen- Pennsylvania, Philadelphia, Pennsylvania; 3Castleman Disease Collaborative Network, ters, constitutional symptoms, and multi-organ failure. Patients can experience thrombocytopenia, Boston, Massachusetts; 4University of anasarca, reticulin fibrosis, renal dysfunction, organomegaly, and normal immunoglobulin levels, – British Columbia, Vancouver, Canada; iMCD-TAFRO. Others experience thrombocytosis, milder effusions, and hypergammaglobulinemia, 5University of Arkansas for Medical – Sciences, Little Rock, Arkansas iMCD-Not Otherwise Specified (iMCD-NOS). Though the etiology is unknown in both subtypes, iMCD symptoms and disease progression are believed to be driven by a cytokine storm, often Correspondence including interleukin-6 (IL-6). However, approximately two-thirds of patients do not respond to David C. Fajgenbaum, Hospital of the anti-IL-6 therapy; alternative drivers and signaling pathways are not known for anti-IL-6 nonres- University of Pennsylvania, 3400 Spruce St, S05094 Silverstein, Division of ponders. To identify potential mediators of iMCD pathogenesis, we quantified 1129 proteins in 13 Translational Medicine and Human plasma samples from six iMCD patients during flare and remission. The acute phase reactant NPS- Genetics, Philadelphia, PA, 19104. PLA2 was the only significantly increased protein (P 5 .017); chemokines and complement were Email: [email protected] significantly enriched pathways. Chemokines represented the greatest proportion of upregulated cytokines, suggesting that iMCD involves a chemokine storm. The chemokine CXCL13, which is essential in homing B cells to germinal centers, was the most upregulated cytokine across all patients (log2 fold-change 5 3.22). Expression of CXCL13 was also significantly increased in iMCD lymph node germinal centers compared to controls in a stromal meshwork pattern. We observed distinct proteomic profiles between the two iMCD-TAFRO patients, who both failed anti-IL-6- therapy, and the four iMCD-NOS patients, in whom all three treated with anti-IL-6-therapy responded, suggesting that differing mechanisms may exist. This study reveals proteomic differences between flare and remission and the potential to molecularly define iMCD subgroups.

1 | INTRODUCTION Recently, a new clinical subtype of iMCD has been described involving thrombocytopenia, anasarca, fibrosis of bone marrow, renal Human Herpesvirus (HHV)-8-negative, idiopathic multicentric Castle- dysfunction, organomegaly (iMCD-TAFRO), and normal immunoglobu- man disease (iMCD) is a polyclonal lymphoproliferative disorder with lin levels.6,7 Other iMCD patients, who do not have TAFRO features an unknown etiology. Approximately 1000 individuals of all ages are (herein referred to as iMCD-Not Otherwise Specified or iMCD-NOS), diagnosed with iMCD each year in the USA,1 and 35%-45% of patients more often demonstrate thrombocytosis, hypergammaglobulinemia, die within 5 years of diagnosis.2–4 Patients experience heterogeneous and less severe fluid accumulation.8 clinical and laboratory abnormalities including constitutional symptoms, Similar clinical and histopathological changes that occur in iMCD anemia, anasarca, renal failure, hypoalbuminemia, thrombocytopenia or also occur in HHV-8-positive MCD in which the HHV-8 virus replicates thrombocytosis, and multicentric lymphadenopathy. Characteristic his- in lymph node plasmablasts and initiates a cytokine storm driving poly- topathological features of the enlarged lymph nodes include dysmor- clonal lymphoproliferation and systemic symptoms.9 Likewise, a cyto- phic (atrophic or hyperplastic) germinal centers, hypervascularization, kine storm—the uncontrolled release of proinflammatory cytokines10— polyclonal lymphoproliferation, and/or polytypic plasmacytosis.5 is presumed to drive iMCD pathogenesis. However, the etiology,

Am J Hematol. 2018;1–11. wileyonlinelibrary.com/journal/ajh VC 2018 Wiley Periodicals, Inc. | 1 2 | AJH PIERSON ET AL. signaling pathways, dysregulated cell types, and composition of the IL-6-therapy (siltuximab or tocilizumab) during their disease course cytokine storm, including the various families of cytokines, such as (Supporting Information Table S2). Of those treated, patients 1–3 interleukins and chemokines, are not known in iMCD. Chemokines reg- responded to treatment and patients 5–6 did not respond. At the time ulate leukocyte migration, lymphoid tissue organization, innate and of flare sample collection, patients 5 and 6 (flare 1) were on anti-IL-6 adaptive immunity, angiogenesis, and immune system development, therapy, and at the time of remission sample collection, patients 1, 2, 3, whereas interleukins are often involved in stimulating immune and 5 were on anti-IL-6 therapy. Complete treatment information is responses, such as inflammation. listed in Supporting Information Table S2. Plasma was isolated follow- Several lines of evidence suggest that excess interleukin-6 (IL-6), a ing standard protocols, stored at 2808C, and shipped in 150 mL ali- cytokine that promotes B cell and plasma cell growth, plays an impor- quots overnight on dry-ice to SomaLogic, Inc (Boulder, Colorado) for tant role in symptoms and disease progression in many iMCD analysis. patients.11,12 Anti-IL-6 monoclonal antibodies (mAb) tocilizumab (anti- IL-6 receptor) and siltuximab (anti-IL-6) have been approved for iMCD 2.2 | High-throughput protein quantification and 5,13,14 in various regions of the world. However, approximately two- analysis thirds of iMCD patients did not respond to IL-6 blockade with siltuxi- ’ mab in the registrational trial,14 and, of the 35 siltuximab nonrespond- SomaLogic s SOMAscan®, a multiplexed, aptamer-based binding-assay, 17 ers in the study, more than one-half had normal IL-6 levels during which has been described in depth previously, was used to quantify active disease (<5pg/mL).12 In agreement, a recent study of serum the relative levels of 1129 proteins. Briefly, the SOMAscan assay uses cytokines found that more than one-half of the 17 iMCD patients Slow-Off-rate Modified Aptamer (SOMAmer) reagents, which are examined had undetectable IL-6 levels during flare.15 These results sug- chemically modified nucleotides, to bind and quantify target proteins in gest that IL-6 independent pathways may be driving disease pathoge- relative fluorescent units directly proportional to the amount of target nesis in a portion of iMCD patients. Current treatment options for protein in the sample. 18 these patients include corticosteroids, rituximab, and cytotoxic chemo- Data analyses were performed using R version 3.4.2. Data were therapies, which have limited efficacy and significant toxicity in normalized by log2 transformation. Fold-change was calculated as the iMCD.16 Efforts to identify new treatment strategies for patients who log2 (flare/remission) for each patient. Both of the patient 6 flare sam- do not respond to anti-IL-6 therapy are limited by a poor understand- ples, labeled as 6.1 and 6.2, were compared to the remission sample. ing of iMCD pathogenesis. Protein function was assigned according to the UniProt molecular func- Here, we performed for the first time unbiased, systematic quanti- tion annotation. For comparison of flare and remission and of clinical fication of plasma proteins in six iMCD patients’ matched flare and subgroups, P values were calculated using Linear Models for Microarray 19 remission samples using SomaLogic SOMAscan® to generate insights and RNA-Seq data (LIMMA) and adjustment for multiple-hypothesis into iMCD pathogenesis. testing was performed using the method of Benjamini and Hochberg (false discovery rate (FDR) < .05).20 2 | METHODS Gene-set enrichment analyses based on the Reactome Pathway Database and Kyoto Encyclopedia of Genes and Genomes (KEGG) 2.1 | Patients and sample collection database and drug enrichment analysis based on the Library of Inte- grated Network-Based Cellular Signatures (LINCS) 1000 database were All patients consented to the research, which was approved by the Uni- performed using Enrichr.21,22 The LINCS 1000 database was used to versity of Arkansas for Medical Sciences Institutional Review Board. identify compounds that decrease gene expression of upregulated pro- Clinical data were collected as part of routine care. Between 2007 and teins. Ingenuity® Pathway Analysis (IPA) software was used for gene- 2016, peripheral blood samples were collected from six patients whose set enrichment and pathway analyses. Enriched gene-sets identified diagnoses are consistent with the iMCD diagnostic criteria.5 One sam- among upregulated and/or downregulated proteins were compared ple during flare (defined as C-Reactive Protein (CRP) > 10 mg/L and against enriched gene-sets identified among the background of 1129 the presence of at least two iMCD-related minor diagnostic criteria5) quantified proteins for all gene-set enrichment analyses. When appro- and one sample during remission (defined as CRP < 10 mg/L and less priate, fold-enrichment ratio and a one-sided Fisher’sExactP value than two minor diagnostic criteria5) were collected from each patient. were calculated for comparison. A P value < .05 was considered signifi- An additional sample was collected for patient 6 during a subsequent cant, unless otherwise noted. A combined score, calculated by the log flare, which occurred three years after the first flare. In total, 13 sam- (P value)*z-score was used to rank the most significant compounds. ples collected from six iMCD patients were analyzed. Data analysis was performed by SKP, AJS, DS, AS, YR, HL, and DCF, The demographic and disease characteristics of the six iMCD and all authors had access to the primary clinical data. patients are presented in Supporting Information Table S1. Median age at diagnosis was 44.5 years, similar to previous reports.16 One patient 2.3 | Immunohistochemistry is female (16.7%). Five of six patients are white (83.3%), and one patient is black (16.7%). Patients 1–4 are iMCD-NOS, whereas patients Immunohistochemistry (IHC) staining of formalin fixed paraffin embed- 5–6 are iMCD-TAFRO. Five of the six patients were treated with anti- ded lymph node tissue from each iMCD patient and five sentinel lymph PIERSON ET AL. AJH | 3 nodes from breast cancer patients with no evidence of metastasis was 3.2 | Pathways involving cytokines, chemokines, and performed at the Pathology Core at the Children’sHospitalofPhiladel- complement are the most enriched during iMCD flare phia. Slides were generated at 5mm thickness. Epitope retrieval was As only one individual protein reached statistical significance across all done for 20 minutes with E2 retrieval solution (Leica Biosystems). IHC iMCD patients when comparing flare to remission samples, we turned was performed on a Leica Bond Max automated staining system (Leica Biosystems) using the Bond Intense R staining kit (Leica Biosystems to gene-set enrichment analyses to identify gene-sets that are signifi- DS9263). Polyclonal rabbit anti-CXCL13 (AF801, R&D Systems) was cantly altered during flare. Using KEGG gene-set database, we analyzed used at a 1:500 dilution and an extended incubation time of 1 hour at all proteins with at least a 2-fold mean increase across all patients during > room temperature. Avidin Biotin Blocking was added (Vector Labs SP- flare (list of 2-fold-change proteins available in Supporting Information 2001) and a Peptide Blocking step was included (DAKO X0909). Slides Table S3). This analysis identified 14 significant KEGG gene-sets; cyto- were digitally scanned at 203 magnification on an Aperio ScanSope kine-cytokine receptor interaction, chemokine signaling pathway, and CS-O slide scanner (Leica Biosystems) and analyzed offline using complement and coagulation cascades were the top 3 gene-sets by Aperio ImageScope and Image Analysis Toolkit software (color decon- combined score. To address the possibility that the set of 1129 proteins volution v9 algorithm). quantified through SOMAscan® was itself enriched for cytokine and chemokine pathways, we compared the results to the background set of all SOMAscan® proteins quantified. Chemokine signaling pathway and | 3 RESULTS complement and coagulation cascades remained significantly enriched (P 5 .03 and P 5 .04, respectively), and the P value for cytokine-cytokine 3.1 | iMCD flare and remission samples have distinct receptor interaction was .09 (Supporting Information Table S4), which proteomic profiles confirmed enrichment of these pathways among the most upregulated To visualize patterns in proteomic alterations that occur during iMCD proteins. We repeated gene-set analyses using the Reactome database. flares, we performed principal component analysis (PCA) on all 13 sam- Gene-sets enriched at least 2-fold relative to the background set of all ples (Figure 1A). The first principal component (PC1) did not clearly SOMAscan® proteins included cytokine signaling, such as G-protein separate based on clinical subtype or disease state. The second princi- coupled receptors and IL-1, 24, 210, and 213 signaling, chemokine sig- pal component (PC2), which explained 13.9% of variance, separated naling, and the complement cascade (Supporting Information Table S5). flare and remission samples. Six of seven flare samples demonstrated a Results from both databases are consistent with involvement of cyto- greater value on the PC2 component axis than all remission samples, kine and chemokine signaling and complement activation in iMCD flare. and each flare sample demonstrated a greater PC2 value than its respective remission sample. To better understand the proteomic alter- 3.3 | Chemokines are the most upregulated family of ations that occur during flare, we plotted the 25 proteins contributing cytokines during iMCD flare the most to PC2 based on absolute value of the features in the loading Given the enrichment of cytokine signaling and the generally accepted vector (Figure 1B). Hemoglobin and Serum Amyloid A (SAA) contrib- model of iMCD as a cytokine storm,23 we performed additional analyses uted the most to PC2. Hemoglobin was the most downregulated pro- to investigate which cytokines were most elevated. Of the 120 cyto- tein across all patients, which is consistent with anemia that occurs during flare. SAA, a cytokine-induced acute phase reactant produced kines measured by the SOMAscan® platform the average fold-change by the liver, was the upregulated protein that contributed the most to of chemokines was significantly higher than interleukins (P 5 .048) and differentiating flare from remission. Other acute phase reactants, such other cytokines (P 5 .025) (Figure 2A, Supporting Information Table S6). as Haptoglobin, CRP, Non-Pancreatic Secretory Phospholipase A2 Chemokines represented six of the eight cytokines that exceeded 2-fold (NPS-PLA2), and Complement 3b (C3b) were also among the top 10 upregulation. Further, nearly one-half of the quantified chemokines strongest contributors to PC2. Five of the remaining top 25 contribut- (43.6%) were represented among the top quartile, which was signifi- ing proteins demonstrate cytokine or chemokine activity, including Tis- cantly greater than interleukins (11.1%, P 5 .006) or other cytokines sue Inhibitor of Metalloproteases-1 (TIMP-1), chemokine C-X-C motif (18.5%, P 5 .008). The three chemokines that were most upregulated chemokine ligand 13 (CXCL13), C-C motif ligand (CCL) 23 (CCL23), during flare were CXCL13 (or B Lymphocyte Chemoattractant (BLC)), CCL21, and CCL14. CCL23 (or Myeloid Progenitor Inhibitory Factor 1-1 (MPIF-1)), and To visualize proteins significantly different in flare compared to CCL21 (or 6CKine). In contrast to the increased abundance of chemo- remission across all iMCD patients, we performed a LIMMA modeling kines, relatively few interleukins were upregulated during iMCD flare. analysis of the log2 fold-change of all proteins and produced a volcano IL-6 was the most upregulated interleukin on average. However, this plot of the results (Figure 1C). However, only one protein, the acute result is confounded by patients receiving anti-IL-6-therapy, which is phase reactant, NPS-PLA2, reached statistical significance after adjust- known to increase circulating IL-6 levels.24 The highest flare IL-6 level ment when compared across all iMCD patients (P 5 .017). In summary, was found in the only patient (patient 5) who was receiving anti-IL-6- iMCD flare and remission samples have distinct proteomic profiles therapy during the flare draw, and the highest remission IL-6 levels were involving consistent trends in acute phase reactants, cytokines, and found in the patients on anti-IL-6-therapy. Given the potential bias of chemokines between flare and remission. anti-IL-6 therapy, we removed samples obtained during anti-IL-6- 4 | AJH PIERSON ET AL.

FIGURE 1 Proteomic alterations during iMCD flare. (A) Principal component analysis of all iMCD samples (N 5 13). PC2, which accounted for 13.9% of variance, separated flare samples from remission. (B) Log2 fold-change of the top 25 proteins contributing to PC2; top contrib- utory proteins consist of acute phase reactants, cytokines, and chemokines. (C) Volcano plot showing the log2 fold-change against the 2log10 adjusted P value following LIMMA analysis. One protein (NPS-PLA2) reached statistical significance (horizontal line) following adjustment. Proteins in blue have >2-fold-change (log2 fold-change of 21.0 and 1.0, delineated by vertical lines) or are significantly elevated (2log10 (Adjusted P value) > 1.301, delineated by horizontal line) between flare and remission. Fifteen select proteins are identified by name. (Abbreviations: BSP: Bone sialoprotein 2; C3b: Complement C3b; CCL14: C-C motif chemokine 14; CK-MM: Creatine kinase M- type; CRP: C-reactive protein; HMG-1: High mobility group protein B1; LBP: Lipopolysaccharide-binding protein; NPS-PLA2: Phospholipase A2, membrane associated; SAA: Serum amyloid A-1 protein; FDR: False discovery rate) [Color figure can be viewed at wileyonlinelibrary.com] therapy and calculated a log2-fold-change for each of the remaining node tissue from each patient prior to iMCD treatment. Compared flare samples (patients 1, 2, 3, 4, 6 (flare 2)) relative to the average of the to sentinel lymph node tissue from breast cancer patients without remaining remission samples (patients 4, 6). Patients 1 and 2 had greater evidence of metastasis (control), germinal centers in iMCD tissue than 3-fold increased IL-6, while patients 3, 4, and 6 had less than 50% demonstrated significantly increased medium (P 5 .047) and strong change. These results suggest that the cytokine storm observed in (P 5 .022) staining and significantly decreased weak staining iMCD includes significant upregulation of chemokines. (P 5 .047) (Figure 2B-D). CXCL13 reactivity appears in both a diffuse, meshwork pattern and a punctate pattern in the germinal cen- 3.4 | CXCL13 expression is elevated in iMCD lymph ter region, likely representing follicular dendritic cells (FDCs) and T node tissue follicular helper (TFh) cells, respectively. These findings suggest that To identify a potential cellular source of the elevated circulating lymph node tissue may be the source of elevated levels of circulat- CXCL13, we performed immunohistochemistry on iMCD lymph ing CXCL13 during iMCD flare. PIERSON ET AL. AJH | 5

FIGURE 2 Chemokines are the most upregulated type of cytokine, and the most upregulated plasma chemokine, CXCL13, demonstrates elevated expression in iMCD lymph node tissue. (A) Bar plot showing mean log2 fold-change of all chemokines (N 5 39), interleukins (N 5 27), and other cytokines (N 5 54) quantified by SOMAscan®. Mean fold-change among chemokines was significantly greater than that of interleukins or other cytokines (P 5 .048 and P 5 .025, respectively). CXCL13 was the most upregulated cytokine (log2 fold-change 3.22), followed by CCL23 (log2 fold-change 1.71) and CCL21 (log2 fold-change 1.57). IL-6 was the top upregulated interleukin (log2 fold-change 0.65). (B) High power image of a representative germinal center from a sentinel lymph node from a breast cancer patient with no evidence of metastasis (Control); (C) high power image of a representative germinal center from an iMCD lymph node, which was resected prior to receiving any iMCD therapy (iMCD) (bar 5 100 mm). Tissue was immunostained with an antibody against CXCL13. (D) Quantification of germinal center CXCL13 staining intensity, as the percentage of pixels, for six iMCD (orange) and five control (green) lymph nodes. Aperio ImageScope was used for all analyses. (aCXCL13; bCCL23; cCCL21; dIL-6; *P < .05) [Color figure can be viewed at wileyonlinelibrary.com]

3.5 | Bioinformatic analyses reveal distinct proteomic subtype (iMCD-TAFRO vs iMCD-NOS) and response to anti-IL-6- profiles between groups associated with clinical therapy. Both iMCD-TAFRO cases did not respond to anti-IL-6- subtype and anti-IL-6 response therapy whereas the three iMCD-NOS patients treated with anti-IL-6- therapy responded (one iMCD-NOS patient never received anti-IL-6- To investigate whether patterns of proteomic alteration differ across therapy). We next analyzed correlations between samples by compar- patients, we performed a PCA of the log2 fold-change values for all ing all 1129 proteins (Figure 3B). While there was positive correlation proteins for each flare-remission pair (Figure 3A). Strikingly, PC1 across the proteomes within the iMCD-TAFRO and iMCD-NOS (accounting for 62.7% of the variance) separated patients by clinical groups, the proteomes of iMCD-TAFRO patients were negatively 6 | AJH PIERSON ET AL.

FIGURE 3 Two iMCD subgroups demonstrate unique proteomic profiles. (A) Principal component analysis of paired iMCD samples identifies two subgroups of patients by PC1 (62.7% of variance) that separate based on clinical subtype and response to anti-IL-6-therapy. (B) Pearson correlation plot shows positive correlation within the subgroups and negative correlation between the subgroups. (C). Heat map of all 1129 proteins demonstrates anti-correlation of many proteins; patients clustered together according to subgroup [Color figure can be viewed at wileyonlinelibrary.com] correlated with the iMCD-NOS patients. We then visualized the rela- 3.6 | PI3K/Akt/mTOR identified as a potential tive values of all proteins in a heat map with unsupervised hierarchical targetable pathway in iMCD clustering of paired samples (Figure 3C). iMCD-TAFRO and iMCD- To generate insights into potential signaling pathways, we performed NOS patients were separated on the first tier. Many of the most upreg- gene-set and pathway analyses on the proteins with >2-fold-change ulated proteins in iMCD-TAFRO patients were downregulated in separately for iMCD-TAFRO and iMCD-NOS groups (listing of proteins iMCD-NOS and vice versa (Supporting Information Table S7). Further, in Supporting Information Tables S8 and S9). To ensure that our results we performed a LIMMA modeling analysis on all proteins for iMCD- were not a product of enrichment among the 1129 analytes that were TAFRO and iMCD-NOS subgroups and plotted the results. Though no quantified, we compared results to the background of all SOMAscan® proteins reached statistical significance in the iMCD-NOS subgroup, proteins. IPA ranked “Phosphoinositide 3-kinase (PI3K) signaling in B 229 proteins were statistically significant after adjustment in the lymphocytes” in the top two canonical pathways for both clinical sub- iMCD-TAFRO subgroup (Supporting Information Figure S1). These groups by P value. All of the top five ranked pathways for iMCD- results demonstrate clear differences in proteomic profiles between TAFRO and iMCD-NOS were significant by P < .001 (Supporting Infor- groups of iMCD patients. However, it is unclear if this distinction is mation Table S10). When KEGG gene-set analyses were performed on due to clinical subtypes and/or anti-IL-6 response. PIERSON ET AL. AJH | 7

TABLE 1 Top 20 nonduplicated compounds identified to downreg- proteins. The identification of the PI3K/Akt/mTOR pathway across ulate proteins that are >2-fold upregulated during iMCD-TAFRO analyses performed on four separate gene-set databases suggest that it flare may be a novel therapeutic target for treating iMCD. Compound Combined name Class of drug score 4 | DISCUSSION GDC-0980 PI3K & mTOR inhibitor 47.5

GSK-1059615 PI3K & mTOR inhibitor 46.8 Here, we have systematically characterized in a small patient cohort the breadth of soluble protein changes during iMCD flare in peripheral XMD-1150 Selective LRRK2 inhibitor 45.1 blood. Our results support the general model that iMCD flares involve Neratinib TKI (EGFR/HER2) 43.2 a cytokine storm and acute inflammation. Acute phase reactants con- Nintedanib TKI (VEGFR/PDGFR) 41.4 tributed strongly to the altered proteomic profile of patients in flare. BIX-01294 Histone methyl transferase inhibitor 37.3 NPS-PLA2, the only protein to meet statistical significance across all

KIN001–244 PDK1 inhibitor 37.2 samples, has been found to be elevated in malignancies and inflammatory disorders and highly correlated with other inflammatory markers, I-BET151 BET family inhibitor 37.0 including CRP and SAA.25–28 Gefitinib TKI (EGFR) 36.7 While iMCD has been described as an IL-6-driven lymphoprolifera- Trametinib MEK inhibitor 36.0 tive disorder,29 recent findings suggest that IL-6 is not the driving cyto- XMD-892 BMK1/ERK5 inhibitor 35.3 kine in all cases, and >50% of patients with iMCD do not respond to anti-IL6-therapy.14 Our findings further indicate that iMCD pathogene- AZD-6482 PI3K inhibitor 35.3 sis is more complex than previously described. In our cohort, anti-IL-6- Geldanamycin Benzoquinone Ansamycin Antibiotic 34.9 therapy induced a complete response in three of five patients in whom Celastrol Triterpenoid antioxidant 34.7 it was administered. However, the flare IL-6 levels were not highly Buparlisib PI3K inhibitor 34.4 upregulated compared to paired remission samples in these patients.

SAR-245408 PI3K inhibitor 34.1 When the anti-IL-6 treated samples were removed, we found that the degree of fold-change may have been underestimated. It is also possi- GSK-1070916 Aurora B/C kinase inhibitor 33.2 ble that some patients may have dysregulation in IL-6 signaling such Dasatinib TKI (BCR/ABL and Src) 32.9 that small changes in IL-6 levels result in major downstream effects.30 SB-239063 MAPK Inhibitor 32.8 We previously found a significantly increased frequency of an IL-6 NVP-BEZ235 PI3K & mTOR inhibitor 32.4 receptor polymorphism among iMCD patients (49%) compared to controls (33%). Individuals with this polymorphism expressed significantly Proteins upregulated > 2-fold during iMCD-TAFRO flare were analyzed using the LINCS 1000 database. Six of the top 20 compounds identified higher levels of soluble IL-6 receptor, which we hypothesized could to downregulate the upregulated proteins are of the PI3K & mTOR contribute to increased IL-6 activity through trans-signaling.31 inhibitor class. A major finding in this study is that chemokines are significantly the iMCD-TAFRO cases, MAPK signaling and chemokine signaling upregulated compared to interleukins and other cytokines during ‘ ’ were significantly enriched compared to background proteins (P 5 .02 iMCD flare. These data suggest that the cytokine storm described in ‘ ’ and P 5 .02). PI3K-Akt was 1.22-fold enriched and had the third high- iMCD would be more appropriately considered a chemokine storm. est combined score, but did not meet significance when compared to The most upregulated circulating cytokine is the chemokine CXCL13, background proteins (P 5 .16) (Supporting Information Table S11). For which is essential for lymph node development and B cell homing to iMCD-NOS cases, KEGG gene-set analyses identified chemokine sig- germinal centers; upregulated levels of CXCL13 have been observed in naling and Vascular Endothelial Growth Factor (VEGF) signaling among disorders that overlap clinically and histopathologically with 32–36 the top five gene-sets, which were also significantly enriched over iMCD. This finding is also notable given that two flare samples background (all P < .01) (Supporting Information Table S12). Lastly, we were collected from patients (patient 2 and 6, flare 1) while on rituxi- performed an enrichment analysis for compounds in the LINCS 1000 mab, which has been shown to decrease serum CXCL13 levels and database that decrease gene expression of proteins with > 2-fold may have dampened CXCL13 fold-change between flare and – upregulation during flare among iMCD-TAFRO and iMCD-NOS cases. remission.37 39 CXCL13 levels were still more than 4-fold higher in After multiple comparison adjustment, six of the 20 most significant flare compared to remission in those patients. Additionally, no patients compounds (highest combined score) identified to downregulate pro- received rituximab within 12 months of remission sample collection, so tein expression for iMCD-TAFRO patients were dual PI3K/mechanistic decreased remission sample CXCL13 is not expected. Thus, CXCL13 target of rapamycin (mTOR) inhibitors or PI3K inhibitors. The remaining may have been further upregulated across patients if rituximab had not 14 compounds each had different targets (Table 1). For iMCD-NOS been administered during two patients’ flare samples. patients, no compounds were identified below the false discovery rate Consistent with proteomic data, we found increased germinal cen- to downregulate protein expression of the >2-fold upregulated ter CXCL13 staining in a diffuse, meshwork pattern in iMCD patients’ 8 | AJH PIERSON ET AL. lymph node tissue compared with metastasis-free sentinel lymph nodes controls. IL-10 and IL-23 were upregulated in iMCD-TAFRO in our from breast cancer patients. Lymph node tissue from all six iMCD study, but demonstrated minimal magnitude and inconsistent changes patients was collected at the time of diagnosis prior to any treatment in iMCD-NOS. VEGF-A (VEGF-121) was greater than two-fold upregu- administration, and therefore the observed CXCL13 is representative lated on average across all patients in our study. Elevated VEGF-A, of the treatment-naive state of iMCD flare. CXCL13 is primarily which promotes cell survival, angiogenesis, and vascular permeability, secreted by FDCs and Tfh cells to home B cells into lymph node germi- has been previously described in iMCD,16,49 and patients often demon- nal centers (via CXCR5).40 Diffuse, strong expression of CXCL13 on strate VEGF-related symptoms, such as highly vascularized lymph FDCs has been described in lymph node tissue from unicentric Castle- nodes, eruptive cherry hemangiomatosis,50 and vascular leak syn- man disease and HHV-8-positive MCD.41 FDC prominence is fre- drome.5 Interestingly, IL-6 was not significantly increased in either the quently observed in iMCD,5 and FDC dysplasia has been occasionally iMCD-TAFRO or iMCD-NOS groups compared to healthy controls in reported.41–45 Similar to what is observed in iMCD, increased numbers the Iwaki study. of lymphoid follicles are seen in CXCL13 overexpressing mice.32,46 The one chemokine measured by Iwaki et al, CXCL10 (or IFN- However, the germinal centers in iMCD are typically small with few B g-Inducible Protein 10 (IP-10), was the only cytokine significantly cells and atrophic features, which is observed in CXCL13 knock-out higher in iMCD-TAFRO flare compared with both iMCD-NOS flare and mice. Overlap between the phenotype observed in these mice and the healthy controls. Though CXCL10 was only the 18th most-upregulated histopathological hallmarks of iMCD47 suggest that there may be dys- cytokine on average across patients in our study, iMCD-TAFRO regulation of CXCL13 signaling in iMCD. Perhaps elevated CXCL13 patients demonstrated >50% upregulation of CXCL10. Like CCL23, secreted by FDCs could drive B cell maturation into plasma cells, which which was highly upregulated in our study, CXCL10 plays an important are increased in iMCD lymph nodes,5 and could lead to auto-antibody role in the recruitment of T cells to sites of inflammation. producing plasma cells. Alternatively, the increased CXCL13 could rep- Iwaki et al. also found that platelet-derived growth factor subunit resent increased, nonspecific germinal center activity,40 which has A (PDGF-AA), which is released upon platelet activation,15,51 was the been observed in related disorders,32–36 or a normal response to the only cytokine significantly lower in iMCD-TAFRO compared with con- failure of B cells to populate the atrophic germinal centers. trols. There was also a trend toward decreased PDGF-AA in iMCD- CCL21 and CCL23 are also highly upregulated during iMCD flares TAFRO compared to iMCD-NOS (P 5 .09).15 PDGF-AA was and have functions directly relevant to lymphoid cell organization. Like likewise > 2-fold downregulated in iMCD-TAFRO and >2-fold upregu- CXCL13, CCL21 is a homeostatic chemokine produced by fibroblastic lated in iMCD-NOS in our study. PDGF-AA has been found to be ele- reticular cells, a type of lymph node stromal cell, that attracts dendritic vated in immune/idiopathic thrombocytopenic purpura (ITP), in whom cells and T cells to the T cell zone/interfollicular space and is essential the thrombocytopenia is caused by peripheral thrombocyte consump- for normal lymph node morphology, which is perturbed in iMCD. tion.52 The etiology of the thrombocytopenia in iMCD-TAFRO is not CCL23 attracts T cells and monocytes to sites of inflammation and pro- known. Though iMCD-TAFRO and ITP share similar clinical features, motes angiogenesis (via CCR1). Our results suggest that chemokine including thrombocytopenia and megakaryocyte hyperplasia, the diver- upregulation in lymph node stromal cells—and considering the mesh- gent PDGF-AA levels suggest that there may be different mechanisms work staining pattern, FDC stromal cells, in particular—may contribute involved in the thrombocytopenia in these disorders.15 Despite differ- to iMCD pathogenesis and warrant further investigation. ences in assays, sample types, and comparison groups, there were sev- Unexpectedly, we saw strong increases in the abundance of com- eral consistent findings between these studies, such as significant plement proteins during iMCD flares. Increases in complement factors differences between iMCD-TAFRO and iMCD-NOS. have not been previously described in iMCD. C3b, the main effector We observed striking differences between patient proteomes from molecule of the complement system, and its cleavage products C3a the two groups that associated with iMCD-TAFRO and iMCD-NOS. In and C3a-des-arginine were among the most upregulated proteins in our study, these groups also associated with response to anti-IL6- the iMCD-TAFRO cases. Auto-antibodies, which are present in approx- therapy. Notably, over 200 proteins were significantly upregulated or imately 30% of iMCD cases, can trigger complement activation when downregulated after adjustment in the iMCD-TAFRO group, which bound to self-antigen.16 Future studies are required to determine com- would be expected considering the high correlation between patients, plement’s role in iMCD pathogenesis and will be an important area to while no proteins in the iMCD-NOS group met significance. Both explore given there is already a safe and effective anti-complement iMCD-TAFRO patients were anti-IL-6 nonresponders and all three antibody in clinical use.48 iMCD-NOS patients treated with anti-IL-6-therapy responded. Impor- Recently, Iwaki et al. quantified the levels of 18 serum cytokines tantly, there have been reports of iMCD-TAFRO cases that have from 11 iMCD-TAFRO patients in flare, 6 iMCD-NOS patients in flare, responded to anti-IL-6-therapy53–59 and others that have not.50,60–66 and 21 healthy controls.15 Among the 18 cytokines measured by fluo- The proteomic differences between the two groups may represent dif- rescent bead immunoassay, there were 13 interleukins, 1 chemokine, ferences between clinical subtypes or be a signature of anti-IL-6 and 4 other cytokines. All but one cytokine (IL-9) were among the response. In either case, the most upregulated proteins and enriched 1129 analytes quantified in our study, and CXCL13 was not measured signaling pathways may constitute therapeutic targets. by Iwaki et al. They found significantly increased IL-10, IL-23, and PI3K, a key member of the PI3K/Akt/mTOR signaling pathway, VEGF-A in both iMCD-TAFRO and iMCD-NOS compared to healthy was identified as a top canonical pathway in both subgroups, and six PIERSON ET AL. AJH | 9 inhibitors of PI3K/Akt/mTOR signaling were identified in the top disease mediators will be critical to building a better understanding of twenty compounds predicted to downregulate expression of upregu- iMCD and leading the field to better treatments. lated proteins in iMCD-TAFRO patients. PI3K/Akt/mTOR is a key reg- ulator of angiogenesis, lymphoproliferation, and inflammation that is therapeutically targetable. Results suggesting involvement of the PI3K/ ACKNOWLEDGMENTS Akt/mTOR pathway were consistently significant across the different We wish to thank Kathleen Sullivan and Melanie Ruffner for assis- pathway and gene-set databases, strengthening our finding that the tance with proteomics pathway analysis. We wish to thank Amy PI3K/Akt/mTOR signaling pathway may be a rational treatment target Greenway, Jossie Carreras-Tartak, Michael Croglio, and Michael Guo for anti-IL-6 nonresponders. for their contributions to the success of this study. We wish to A major limitation of this study is the small number of patient sam- thank the Jayanthan Family and the Castleman Disease Collaborative ples analyzed. Though our decision to include only iMCD patients with Network/Castleman’s Awareness & Research Effort for funding. matched flare and remission samples limited the number of eligible patients with this rare disease, we wanted to identify proteomic changes that can be attributed to disease state and therapies. We ORCID expect that the continued growth of the international, multi-center Castleman Disease Collaborative Network,67 which was started in David C. Fajgenbaum http://orcid.org/0000-0002-7367-8184 2012, will be instrumental in recruitment of larger numbers of patients with this rare disease for future translational studies. Furthermore, REFERENCES while large-scale plasma protein profiling gives a systematic view of [1] Munshi N, Mehra M, van de Velde H, et al. Use of a claims data- protein changes, this approach may overlook highly potent proteins base to characterize and estimate the incidence rate for Castleman disease. Leuk Lymphoma. 2015;56(5):1252–1260. where small changes in concentration result in major biological effects as well as nonsecreted factors, which were not assayed. Given the [2] Dispenzieri A, Armitage JO, Loe MJ, et al. The clinical spectrum of Castleman’s disease. Am J Hematol. 2012;87(11):997–1002. exploratory nature of the analysis, there is also the potential for false [3] Shin D-Y, Jeon YK, Hong Y-S, et al. Clinical dissection of multicen- discovery. Therefore, we employed a false discovery rate and per- tric Castleman disease. Leuk Lymphoma. 2011;52(8):1517–1522. formed gene-set enrichment analyses to improve our power for identi- [4] Melikyan AL, Egorova EK, Kovrigina ﬀ, et al. Clinical and morpho- fying significantly enriched or depleted groups of genes. Another logical features of different types of Castleman’s disease]. Ter Arkh. limitation of this study is that renal dysfunction affects excretion of cir- 2015;87(7):64–71. [ 68 culating proteins, particularly small (<25 kDa) proteins, and three [5] Fajgenbaum DC, Uldrick TS, Bagg A, et al. International, evidence- patients (2, 5, 6) experienced borderline or elevated creatinine sugges- based consensus diagnostic criteria for HHV-8-negative/idiopathic – tive of renal dysfunction during flare. Decreased excretion may there- multicentric Castleman disease. Blood. 2017;129(12):1646 1657. fore have contributed to increased concentration of some proteins, [6] Iwaki N, Fajgenbaum DC, Nabel CS, et al. Clinicopathologic analysis of TAFRO syndrome demonstrates a distinct subtype of HHV-8- though these increased proteins may also contribute to iMCD patho- negative multicentric Castleman disease. Am J Hematol. 2016;91(2): physiology. Further, treatments, such as anti-IL-6 therapies (siltuximab 220–226. and tocilizumab), rituximab, corticosteroids, and chemotherapies may [7] Masaki Y, Kawabata H, Takai K, et al. Proposed diagnostic criteria, affect the plasma proteome. We addressed the potential effects of disease severity classification and treatment strategy for TAFRO anti-IL-6 therapy and rituximab and provided each of the treatments syndrome, 2015 version. Int J Hematol. 2016;103(6):686–692. administered to these patients. None of the major observations in this [8] Srkalovic G, Marijanovic I, Srkalovic MB, Fajgenbaum DC. TAFRO study segregated to only patients on rituximab, corticosteroids, or che- syndrome: new subtype of idiopathic multicentric castleman disease. – motherapies, and other than those discussed, we are not aware of any Bosn J Basic Med Sci. 2017;17(2):81 84. specific effects of these agents on the noted proteins. Lastly, the lack [9] Uldrick TS, Polizzotto MN, Yarchoan R. Recent advances in Kaposi sarcoma herpesvirus-associated multicentric Castleman disease. Curr of samples from a healthy control population prevent comparison of Opin Oncol. 2012;24(5):495–505. disease state from both flare and remission with otherwise healthy [10] Tisoncik JR, Korth MJ, Simmons CP, et al. Into the eye of the cyto- controls. However, we sought to identify proteomic changes occurring kine storm. Microbiol Mol Biol Rev. 2012;76(1):16–32. within individuals between flare and remission. Comparisons between [11] Yoshizaki K, Matsuda T, Nishimoto N, et al. Pathogenic significance iMCD flare samples and healthy controls can introduce inter-individual of interleukin-6 (IL-6/BSF-2) in Castleman’s disease. Blood. 1988;10 differences in plasma proteomes that can make changes difficult to (4):13–17. interpret. [12] Casper C, Chaturvedi S, Munshi N, et al. Analysis of inflammatory Despite these limitations our study provides insights into the and anemia-related biomarkers in a randomized, double-blind, placebo-controlled study of siltuximab (anti-IL6 monoclonal antibody) pathogenesis of iMCD, including identification of a ‘chemokine storm’ in patients with multicentric castleman disease. Clin Cancer Res. associated with iMCD flares and potential involvement of lymph node 2015;21(19):4294–42304. stromal cells. Additionally, our studies reveal the proteomic differences [13] Nishimoto N, Kanakura Y, Aozasa K, et al. Humanized anti- between flare and remission and the potential to molecularly define interleukin-6 receptor antibody treatment of multicentric Castleman iMCD subgroups. Future studies testing the contribution of individual disease. Blood. 2005;106(8):2627–2632. 10 | AJH PIERSON ET AL.

[14] van Rhee F, Wong RS, Munshi N, et al. Siltuximab for multicentric [32] Klimatcheva E, Pandina T, Reilly C, et al. CXCL13 antibody for the Castleman’s disease: a randomised, double-blind, placebo-controlled treatment of autoimmune disorders. BMC Immunol. 2015;16(1):6–6. – trial. Lancet Oncol. 2014;15(9):966 974. [33] Kramer JM, Klimatcheva E, Rothstein TL. CXCL13 is elevated in [15] Iwaki N, Gion Y, Kondo E, et al. Elevated serum interferon Sjogren€ ’s syndrome in mice and humans and is implicated in disease g-induced protein 10 kDa is associated with TAFRO syndrome. Sci pathogenesis. J Leukoc Biol. 2013;94(5):1079–1089. Rep. 2017;7:42316. (January): [34] Barone F, Bombardieri M, Manzo A, et al. Association of CXCL13 [16] Liu AY, Nabel CS, Finkelman BS, et al. Idiopathic multicentric Cas- and CCL21 expression with the progressive organization of tleman’s disease: a systematic literature review. Lancet Haematol. lymphoid-like structures in Sjogren€ ’s syndrome. Arthritis Rheum. 2016;3(4):e163–e175. 2005;52(6):1773–1784. [17] Gold L, Ayers D, Bertino J, et al. Aptamer-based multiplexed pro- [35] Dunleavy K, Wilson WH, Jaffe ES. Angioimmunoblastic T cell lym- teomic technology for biomarker discovery. PLoS One. 2010;5(12): phoma: pathobiological insights and clinical implications. Curr Opin e15004. Hematol. 2007;14(4):348–353. [18] R Core Team. R: a Language and Environment for Statistical Com- [36] Dupuis J, Boye K, Martin N, et al. Expression of CXCL13 by neo- puting. Vienna, Austria: R Foundation for Statistical Computing; plastic cells in angioimmunoblastic T-cell lymphoma (AITL): a new 2017. diagnostic marker providing evidence that AITL derives from follicu- – [19] Smyth GK, Michaud J, Scott HS. Use of within-array replicate spots lar helper T cells. Am J Surg Pathol. 2006;30(4):490 494. for assessing differential expression in microarray experiments. Bio- [37] Pranzatelli MR, Tate ED, Travelstead AL, Verhulst SJ. Chemokine/ informatics. 2005;21(9):2067–2075. cytokine profiling after rituximab: reciprocal expression of BCA-1/ – [20] Benjamini Y, Hochberg Y. Controlling the false discovery rate: a CXCL13 and BAFF in childhood OMS. Cytokine. 2011;53(3):384 389. practical and powerful approach to multiple testing. J R Stat Soc B. [38] Rosengren S, Wei N, Kalunian KC, Kavanaugh A, Boyle DL. 1995;57(1):289–300. CXCL13: a novel biomarker of B-cell return following rituximab [21] Chen EY, Tan CM, Kou Y, et al. Enrichr: interactive and collabora- treatment and synovitis in patients with rheumatoid arthritis. Rheu- – tive HTML5 gene list enrichment analysis tool. BMC Bioinformatics. matology. 2011;50(3):603 610. 2013;14:128. [39] Alvarez E, Piccio L, Mikesell RJ, et al. Predicting optimal response to [22] Kuleshov MV, Jones MR, Rouillard AD, et al. Enrichr: a comprehen- B-cell depletion with rituximab in multiple sclerosis using CXCL13 sive gene set enrichment analysis web server 2016 update. Nucleic index, magnetic resonance imaging and clinical measures. Acids Res. 2016;44(W1):W90–W97. [40] Havenar-Daughton C, Lindqvist M, Heit A, et al. CXCL13 is a [23] Fajgenbaum DC, Van Rhee F, Nabel CS. HHV-8-negative, idiopathic plasma biomarker of germinal center activity. Proc Natl Acad Sci – multicentric Castleman disease: novel insights into biology, pathoge- USA. 2016;113(10):2702 2707. nesis, and therapy. Blood. 2014;123(19):2924–2933. [41] Vermi W, Lonardi S, Bosisio D, et al. Identification of CXCL13 as a [24] Nishimoto N, Terao K, Mima T, et al. Mechanisms and pathologic new marker for follicular dendritic cell sarcoma. J Pathol. 2008;216 – significances in increase in serum interleukin-6 (IL-6) and soluble IL- (3):356 364. 6 receptor after administration of an anti-IL-6 receptor antibody, [42] Chang K-C, Wang Y-C, Hung L-Y, et al. Monoclonality and cytoge- tocilizumab, in patients with rheumatoid arthritis and Castleman dis- netic abnormalities in hyaline vascular Castleman disease. Mod ease. Blood. 2008;112(10):3959–3964. Pathol. 2014;27(6):823–831. [25] Menschikowski M, Hagelgans A, Schuler U, et al. Plasma levels of [43] Yasuda S, Tanaka K, Ichikawa A, et al. Aggressive TAFRO syndrome phospholipase A2-IIA in patients with different types of malignan- with reversible cardiomyopathy successfully treated with combina- cies: prognosis and association with inflammatory and coagulation tion chemotherapy. Int J Hematol. 2016;104(4):512–518. – biomarkers. Pathol Oncol Res. 2013;19(4):839 846. [44] Inoue M, Ankou M, Hua J, et al. Complete resolution of TAFRO [26] Menschikowski M, Hagelgans A, Fuessel S, et al. Serum amyloid A, syndrome (thrombocytopenia, anasarca, fever, reticulin fibrosis and phospholipase A2-IIA and C-reactive protein as inflammatory bio- organomegaly) after immunosuppressive therapies using corticoste- markers for prostate diseases. Inflamm Res. 2013;62(12):1063– roids and cyclosporin A : a case report. J Clin Exp Hematop. 2013;53 1072. (1):95–99. [27] Sullivan CP, Seidl SE, Rich CB, Raymondjean M, Schreiber BM. [45] Sun X, Chang K-C, Abruzzo LV, et al. Epidermal growth factor

Secretory phospholipase A 2, Group IIA is a novel serum amyloid a receptor expression in follicular dendritic cells: a shared feature of target gene. J Biol Chem. 2010;285(1):565–575. follicular dendritic cell sarcoma and Castleman’s disease. Hum – [28] Uusitalo-Seppälä R, Peuravuori H, Koskinen P, Vahlberg T, Rintala Pathol. 2003;34(9):835 840. EM. Role of plasma bactericidal/permeability-increasing protein, [46] Ansel KM, Ngo VN, Hyman PL, et al. A chemokine-driven positive

group IIA phospholipase A 2, C-reactive protein, and white blood feedback loop organizes lymphoid follicles. Nature. 2000;406(6793): cell count in the early detection of severe sepsis in the emergency 309–314. – department. Scand J Infect Dis. 2012;44(9):697 704. [47] Cronin DMP, Warnke RA. Castleman disease - An update on classi- [29] El-Osta HE, Kurzrock R. Castleman’s disease: from basic mecha- fication and the spectrum of associated lesions. Adv Anat Pathol. nisms to molecular therapeutics. Oncologist. 2011;16(4):497–511. 2009;16(4):236–246. [30] Patel M, Ikeda S, Pilat SR, Kurzrock R. JAK1 genomic alteration [48] Hillmen P, Young NS, Schubert J, et al. The complement inhibitor associated with exceptional response to siltuximab in cutaneous eculizumab in paroxysmal nocturnal hemoglobinuria. N Engl J Med. castleman disease. JAMA Dermatol. 2017;153(5):449–452. 2006;355(12):1233–1243. [31] Stone K, Woods E, Szmania SM, et al. Interleukin-6 receptor poly- [49] Nishi J, Arimura K, Utsunomiya A, et al. Expression of vascular morphism is prevalent in HIV-negative Castleman disease and is endothelial growth factor in sera and lymph nodes of the plasma associated with increased soluble interleukin-6 receptor levels. PLoS cell type of Castleman’s disease. Br J Haematol. 1999;104(3):482– One. 2013;8(1):e54610. 485. PIERSON ET AL. AJH | 11

[50] Fajgenbaum DC, Fajgenbaum D, Rosenbach M, et al. Eruptive cherry [61] Tadokoro A, Kanaji N, Hara T, et al. An Uncharted Constellation: hemangiomatosis associated with multicentric Castleman disease: a TAFRO Syndrome. Am J Med. 2016;129(9):938–941. – case report and diagnostic clue. JAMA Dermatol. 2013;149(2):204 208. [62] Tatekawa S, Umemura K, Fukuyama R, et al. Thalidomide for [51] Kaplan DR, Chao FC, Stiles CD, Antoniades HN, Scher CD. tocilizumab-resistant ascites with TAFRO syndrome. Clin Case Rep. Platelet alpha granules contain a growth factor for fibroblasts. 2015;3(6):472–478. – Blood. 1979;53(6):1043 1052. [63] Simons M, Apor E, Butera JN, Treaba DO. TAFRO syndrome associ- [52] Katoh O, Kimura A, Fujimura K, Kuramoto A. Progression activity ated with EBV and successful triple therapy treatment: case report for platelet-derived growth factor in plasma of patients with idio- and review of the literature. Case Rep Hematol. 2016;2016:1. pathic thrombocytopenic purpura and aplastic anemia. Nippon Ket- [64] Tedesco S, Postacchini L, Manfredi L, et al. Successful treatment of – sueki Gakkai Zasshi. 1989;52(1):113 117. a Caucasian case of multifocal Castleman’s disease with TAFRO [53] Sakai K, Maeda T, Kuriyama A, et al. TAFRO syndrome successfully syndrome with a pathophysiology targeted therapy - a case report. treated with tocilizumab: a case report and systematic review. Mod Exp Hematol Oncol. 2015;4(1):3. – Rheumatol. 2016;1 6. [65] Hiramatsu S, Ohmura K, Tsuji H, et al. Successful treatment by rit- [54] Fujiwara S, Mochinaga H, Nakata H, et al. Successful treatment of uximab in a patient with TAFRO syndrome with cardiomyopathy. TAFRO syndrome, a variant type of multicentric Castleman disease Jpn J Clin Immun. 2016;39(1):64–71. with thrombotic microangiopathy, with anti-IL-6 receptor antibody [66] Tanaka H, Bujo S, Itobayashi E, et al. TAFRO syndrome with a rapid – and steroids. Int J Hematol. 2016;103(6):718 723. fatal course despite corticosteroid and tocilizumab therapy. [55] Sakashita K, Murata K, Inagaki Y, Oota S, Takamori M. An anterior J Hematopathol. 2016;9(4):167–171. mediastinal lesion in TAFRO syndrome showing complete remission [67] Fajgenbaum D, Ruth J, Kelleher D, Rubenstein A. The collaborative after glucocorticoid and tocilizumab therapy. Respirol Case Rep. network approach: a new framework to accelerate Castleman’s dis- 2016;4(5):e00173. ease and other rare disease research. Lancet Haematol. 2016;3(4): [56] Kubokawa I, Yachie A, Hayakawa A, et al. The first report of adoles- e150–e175. cent TAFRO syndrome, a unique clinicopathologic variant of multi- [68] Mehan MR, Ostroff R, Wilcox SK, et al. Highly multplexed proteo- ’ centric Castleman s disease. BMC Pediatr. 2014;14(1):139. mic platform for biomarker discovery, diagnostics, and therapeutics. [57] Kawabata H, Kotani S, Matsumura Y, et al. Successful treatment of Complement Ther. 2013;283–300. a patient with multicentric Castleman’s disease who presented with thrombocytopenia, ascites, renal failure and myelofibrosis using tocilizumab, an anti-interleukin-6 receptor antibody. Intern Med. SUPPORTING INFORMATION 2013;52(13):1503–1507. Additional Supporting Information may be found online in the sup- [58] Coutier F, Meaux Ruault N, Crepin T, et al. A comparison of TAFRO porting information tab for this article. syndrome between Japanese and non-Japanese cases: a case report and literature review. Ann. Hematol. 2018;97:401–407. [59] Fujiki T, Hirasawa S, Watanabe S, Iwamoto S, Ando R. Successful treatment by tocilizumab without steroid in a very severe case of How to cite this article: Pierson SK, Stonestrom AJ, Shilling D, TAFRO syndrome. CEN Case Rep. 2017;6(1):105–110. 0 et al. Plasma proteomics identifies a ‘chemokine storm’ in idio- [60] Konishi Y, Takahashi S, Nishi K, et al. Successful treatment of pathic multicentric Castleman disease. Am J Hematol. TAFRO syndrome, a Variant of Multicentric Castleman’s disease, 2018;00:1–11. https://doi.org/10.1002/ajh.25123 with cyclosporine a: possible pathogenetic contribution of Interleu- kin-2. Tohoku J Exp Med. 2015;236(4):289–295.